Bioreceptor-inspired soft sensor arrays: recent progress towards advancing digital healthcare

Soft eye image-sensing arrays

The human eye comprises a lens that collects light and a hemispherical retina. Dense integration of 100-120 million photoreceptors and rod and cone cells in the retina helps convert incident light to action potentials that are transmitted to the brain through millions of nerve fibres [Figure 2A]^[27,130].

Figure 2. (A) Human eye and retina; (B) exploded, side, and top views of EC-EYE; low-resolution cross-sectional SEM image of hemispherical PAM/nanowires; cross-sectional SEM image of nanowires in PAM; high-resolution transmission electron microscopy image of a single-crystalline perovskite nanowire; and a photograph of the PDMS socket; (C) schematic illustration of measurement setup; and (D) image reconstructed (letter “A”) by EC-EYE and its projection on a flat plane. This figure is quoted with permission from Gu et al.^[27]. EC-EYE: Electrochemical eye; PAM: porous aluminium oxide membranes; PDMS: polydimethylsiloxane.

Gu et al. developed a biomimetic artificial electrochemical eye (EC-EYE) consisting of a hemispherical retina with a high-density perovskite nanowire array acting as the photosensor [Figure 2B]^[27]. An ionic liquid electrolyte mimicking vitreous humour served as the front-side common contact with the nanowires. Freestanding hemispherical porous aluminium oxide membranes (PAM) were used to grow a high-density array of perovskite nanowires inside nanochannels. Nanochannels were selectively opened in the PAM layer by using a focused ion beam. A magnetic field was used to align nickel (Ni) microneedles with three exposed nanowires on the PAM layer. Each Ni microneedle formed a pixel with a lateral size of 1 μm and pitch of 200 μm. Thin tubes filled with liquid metal were used to connect to each Ni microneedle through a PDMS socket. The PDMS socket was fabricated using a 3D-printed hedgehog-shaped mould to pattern a 10 × 10 hole array on the socket. Thin soft tubes filled with eutectic gallium indium liquid metal were used to form liquid-metal wires mimicking human nerve fibres behind the retina. Then, 100 such tubes were inserted into the PDMS socket holes, and the socket was attached to the PAM/nanowire surface to form the photodetector array. The liquid metal tubes were connected directly to a computer-controlled 100 × 1 multiplexer on a printed circuit board (PCB) [Figure 2C]. Several optical patterns (e.g., character “A”) were projected onto the EC-EYE, and the photocurrent of each pixel was measured using a current meter. A personal computer (PC) was used for both processing the current data and controlling the multiplexer. The photocurrent values were converted to a grayscale number between zero and 255 to reconstruct the detected objects. Figure 2D shows the character “A” imaged by the EC-EYE and its projection onto a flat plane. The EC-EYE achieved a maximum photocurrent sensitivity of 303.2 mA/W, and the high-density nanowire arrays provided high image resolution. The EC-EYE was demonstrated as a vision system for humanoid robots that resembled the human eye in appearance and characteristics.

Unlike the EC-EYE, which reconstrued the entire eye, Choi et al. focused on developing a soft high-density image sensor and electrode array (i.e., a closed-loop system) integrated with the human retina, as shown in Figure 3A^[116]. The device consists of a high-density, vertically stacked, and ultra-thin soft molybdenum disulfide (MoS₂)-graphene phototransistor (CurvIS) array and ultra-thin neural-interfacing electrodes (UNE) [Figure 3B]. Each stacked electrode and phototransistor is connected through a flexible PCB (FPCB) coated with soft and thick silicone rubber (i.e., soft FPCB). An external light signal is focused by the lens onto the CurvIS array [Figure 3B]. The output photocurrent generated by each phototransistor in response to the external light is amplified by a trans-impedance amplifier, passed through an inverter, and fed into a micro-controller unit (MCU). Figure 3C shows all these components integrated into the soft FPCB. The MCU measures and processes the amplified signal and generates electrical pulses. These electrical pulses are then applied to the relevant UNE electrode to stimulate the retina [Figure 3D]. Moreover, the FPCB also includes image-processing electronics. Figure 3E shows optical camera images of the CurvIS array, UNE stack, and soft FPCB. To fabricate the UNE, chromium (Cr)/Au electrodes were patterned on a PI-coated silicon dioxide (SiO₂) wafer, followed by the deposition of a thin platinum (Pt) layer and a final PI encapsulation layer. A water-soluble tape was used to detach the UNE electrodes from the SiO₂ wafer and laminate them on the retina. The water-soluble tape was then dissolved. Similarly, the CurvIS phototransistor array included a stack of several layers on a thin PI film spin-coated onto a SiO₂ wafer. The layers included a thin layer of silicon nitride (Si₃N₄) as the transistor substrate, interdigitated graphene source/drain electrodes, an ultrathin photo-absorbing MoS₂ layer coated onto the graphene electrodes, an aluminium oxide (Al₂O₃) dielectric layer, a patterned titanium (Ti)/Au layer as the gate, and a final PI encapsulation layer. A water-soluble tape was used to detach the phototransistor array from the SiO₂ wafer. The tape was then shaped into a truncated icosahedron structure, stacked on the UNE, and laminated on the retina. The water-soluble tape was then dissolved. Figure 3F presents the university logo image captured by the CurvIS array. The phototransistor exhibited superior photoresponsivity than a silicon photodetector. This device could be used as an intraocular retinal prosthesis for patients with retinal degeneration, such as macular degeneration or retinitis pigmentosa.

Figure 3. (A) Schematic illustration of the human ocular structure and the integrated soft optoelectronic device; (B) schematic illustration of the high-density CurvIS array based on the MoS₂-graphene heterostructure; (C) phototransistor stack (bottom) and the stimulation electrode (top) connected through the soft FPCB; (D) schematic illustration of the electronics for detecting external light and stimulation; (E) optical camera image of the CurvIS array and UNE stack connected through the soft FPCB; and (F) university logo image captured by the CurvIS array^[116]. FPCB: Flexible printed circuit board; UNE: ultra-thin neural-interfacing electrodes.

Soft skin sensor arrays and virtual/augmented reality actuator arrays for perception

Some animals, such as sharks, skates, and rays, have electroreceptor network-embedded skins that allow them to locate nearby prey without making physical contact^[131]. Each electroreceptor consists of two components: a soft ion-conducting hydrogel core that receives the electric fields generated by the prey animals and an ion-insulating epithelium that acts as a soft encapsulation layer for the hydrogel core [Figure 4A]^[53,132]. These animals compare the electric field intensity detected by each electroreceptor in the network to perceive the relative position of nearby prey (i.e., spatial perception)^[132].

Figure 4. (A) Network of electroreceptors distributed on a ray’s skin for locating nearby prey; (B) schematic illustration of wearable SAER on a hand; (C) image of wearable SAER attached to the hollow of the hand; (D) a person wearing the SAER to perceive spatial information of a dynamic object; and (E) a person wearing the SAER to perceive spatial information of a static object^[117]. SAER: Soft artificial electroreceptor.

This sensing mechanism inspired Song et al. to develop a soft artificial electroreceptor (SAER), which is depicted in Figure 4B^[117]. The SAER consists of a three-layer stack of 3D-printed conductive hydrogel on dielectric elastomers and an additional top-cover elastomer layer. The SAER attached to the hand of a subject is illustrated in Figure 4C. The middle layer is the sensing layer and consists of an array of four hydrogel electric field receivers. The bottom hydrogel layer that is in contact with the skin acts as a shield layer that blocks electrical noise from the skin of the subject. The top hydrogel layer acts as a shield layer, and it ensures that the voltage signal of the sensing layer electrodes is responsive to the distance of the external object, not its direction. To realise reliable adhesion of the 3D-printed hydrogel on the elastomer, the elastomer surface was first soaked in a solution containing an ultraviolet (UV)-assisted grafting agent. The hydrogel was then directly printed on the UV-treated elastomer by using a digital light processing (DLP) 3D printer. By comparing the intensity of the electric fields sensed by each hydrogel receiver in Figure 4C, the SAER was able to locate the relative position of the object. When wearing the SAER, the wearer receives spatial information about nearby objects in real time through auditory signals of specific frequencies generated for each of the receivers. Figure 4D shows the maximum voltage of the blue receiver corresponding to the auditory signal frequency of 0.5 kHz when the moving ball reaches the hand. By contrast, the maximum induced voltage peak of the green receiver when the hand passes the static ball corresponds to the auditory signal frequency of 1.5 kHz, as depicted in Figure 4E. This artificial electroreceptor allows a human to perceive space, a new sensory modality that does not exist in human beings. This additional sensory mechanism could help blind or visually impaired people, who cannot rely on visual sensations, to perceive space.

Virtual reality (VR) usually involves providing audio and visual feedback to generate a virtual world in which users can interact with virtual objects^[133]. In augmented reality (AR), the virtual and real worlds are combined using advanced electronics and data processing systems^[134]. VR/AR devices are not directly inspired by bioreceptors, but they draw upon the human perceptual and sensory system to create immersive experiences. Figure 5A shows how perception and behaviour are shaped in response to multiple stimuli^[135]. Similarly, the sensors integrated within VR/AR devices can measure physiological and body movements of a person to simulate any sensory modality of the human body. When the stimuli received by the brain match the expectations of the sensory input, the brain is more likely to treat the simulated reality as real (i.e., virtual perception) and increase engagement with the perceived illusion. For instance, Liu et al. developed an array of flexible and miniaturised odour generators (OGs) that can be attached to the skin above the lips^[136]. These OGs were able to act as an olfaction interface wirelessly connected to the 3D virtual world and release a range of odours to enhance the realism of the VR/AR experience.

Figure 5. (A) Performance enhancement in several perceptual and behavioural domains owing to multisensory integration^[135]; (B) Exploded-view schematic illustration of a device with 32 independently controlled haptic actuators, NFC electronics, circuit schematic and images, and exploded-view schematic illustration of a haptic actuator; and (C) illustration of “virtual touch process” and “sense of virtual touch” where a girl touches a screen that displays a video feed of her grandmother, who is wearing an epidermal VR device on her hand and arm. This figure is quoted with permission from Yu et al.^[118]. NFC: Near-field communication; VR: virtual reality.

Compared to the existing VR/AR devices that provide interactive images and sounds, Yu et al. developed a battery-free, wirelessly controlled, and powered haptic VR interface that provided touch sensations to users^[118]. This haptic interface consisted of large arrays of millimetre-scale vibratory actuators embedded in soft and conformal sheets of electronics [Figure 5B]. Multiple haptic interfaces of this type were laminated directly on several locations on the skin and controlled remotely by using a computer system. The control signals from the computer system, such as those provided through a touch screen, activated the actuators to establish VR/AR touch experiences. The interface included a large primary coil that covered the entire perimeter of the device platform to harvest adequate power to operate all the actuators. This power was passed through a linear voltage regulator to provide a fixed voltage to each actuator. Separate small near-field communication (NFC) antennas were used to independently control eight actuators through the relevant general purpose input/output (GP I/O) ports. An integrated circuit (IC) switch was associated with each actuator and transformed the fixed voltage into a square wave. To fabricate the haptic actuators, first, a copper (Cu) coil was sealed between two PDMS layers by using two moulds. A Ni-plated neodymium magnet mounted on a circular PI disk with a semi-circular slit. A PDMS ring and a silicone adhesive were used to bond the PI disk and coil together, as illustrated in Figure 5A. For the electronics part, first, a Cu layer was coated on a PI layer and patterned using photolithography and etching to form the interconnect wires, coil, and antennas. Low-temperature solder joints were formed to create electrical connections between all the other electronic components on the patterned Cu layer. An ultra-low-modulus silicone material was used as an adhesive layer between the cloth and the electronic/haptic platform. A layer of skin-coloured PDMS was used as the top encapsulation layer, which additionally provided reversible adhesion to the skin. This unique structure of the entire system facilitates scaling of the interface for various applications. Figure 5C shows an application of the VR interface. When the interface is placed on a display that shows a video feed of a girl’s grandmother, the girl is able to virtually touch the hand of her grandmother through the interface. This touch on the VR interface activates the relative actuators on the VR interface mounted on the grandmother’s hand, who experiences a haptic sensation with a spatio-temporal touch pattern. Other representative applications include providing tactile feedback in robotic prosthetic devices or haptic engagement in gaming.

Soft sense of touch sensor and physiological signal monitoring arrays

The sensation of touch conveys information about the objects with which we interact and about our interactions with them^[6]. Touch is mediated by somatosensory neurons called mechanoreceptors embedded in the skin that relay signals from the peripheral to the central nervous system. The brain processes this information to adjust our grasping force by engaging muscles and tendons to hold an object without crushing it. The tight coupling between the somatosensory and motor systems defines manual dexterity. The high level of dexterity of the human hand allows humans to perform numerous complex tasks. Therefore, the human hand is often used as the inspiration to develop smart skins for future interactive robotics and wearable health monitoring devices^[55,137].

The human skin consists of the epidermis, dermis, and hypodermis, as shown in Figure 6A^[119,138]. The hairs present within the dermis can sensitively detect environmental physical changes and transmit them to the brain through mechanoreceptors and nerve fibres.

Figure 6. (A) Schematic illustration of the human skin and its hair-sensing mechanism; (B) the schematic illustration of capacitive pressure sensor with MCA as the dielectric layer; and (C) optical images of sensor arrays for pressure detection from different objects of various weights and shapes, and the related pressure distribution readouts. This figure is quoted with permission from Zhou et al.^[119]. Ag NW: Ag nanowire; MCA: micro cilia array; PDMS: polydimethylsiloxane.

Inspired by hairy human skin, Zhou et al. developed a flexible micro cilia array (MCA) layer^[119]. The MCA layer worked as a dielectric layer sandwiched between two layers of Ag nanowire (Ag NW)/PDMS electrodes to form a flexible capacitive pressure sensor, as shown in Figure 6B. The MCA layer was then cut into 0.5 × 0.5 cm² pieces and applied as separate dielectric layers in a flexible 5 × 5 capacitive sensor array. A facile method was used to fabricate the MCA layer. First, a PDMS layer was formed on a glass substrate. Then, a layer composed of a mix of carbonyl iron particles (CIPs), PDMS, and a curing agent was spin-coated onto the cured PDMS membrane. The substrate was then moved to the surface, where the PDMS/CIP layer was magnetised using a permanent neodymium (NdFeB) magnet. This magnetic field caused aggregation of the CIPs into chains and, finally, into the cilia array patterns. The images of this sensor array in the presence of four objects of different weights and two objects having rectangular and triangular shapes are presented in Figure 6C. The detected pressure distributions and magnitudes of the four objects and the detected shapes of the rectangular and triangular objects are depicted in Figure 6C. The sensor exhibited a wide pressure detection range of 0-200 kPa with a low detectable pressure of 2 Pa. This property ensured that the sensor was suitable for a range of applications, such as real-time monitoring of elbow and finger bending, voice monitoring, wrist pulse monitoring, and pressure monitoring of the hind sole during standing, walking, and jumping. The device can be used in various applications, including wearable devices, artificial intelligence, and interactive robotics.

Inspired by mechanoreceptors and synapses^[139] (i.e., the signal transmission junctions between two neurons), Kim et al. developed a stretchable sensory-neuromorphic system (SSNS) [Figure 7A]^[120]. The SSNS consisted of one 5 × 5 capacitive pressure sensor array (i.e., artificial mechanoreceptors) and four arrays (i.e., one artificial neuron), each including 5 × 5 resistive random-access memories (RRAMs) (i.e., an artificial synapse). In addition, they integrated an array of quantum dot light-emitting diodes (QLEDs) mimicking the epidermal photonic actuators that exist on golden tortoise beetles to display the output signals of the pressure sensor array. The usage of intrinsically stretchable printed interconnects between these components stabilised the operation of the system under up to 25% stretching, which is similar to the range of skin deformation. To fabricate the pressure sensor array, a sinter-free ink was prepared by mixing the polymer [PDMS-(4,4′-methylenebis(phenyl isocyanate) (MPU)0.4-isophorone diisocyanate (IU)0.6] with methyl isobutyl ketone (MIBK) and Ag flakes. The ink was screen-printed using a metal screen mask to pattern the bottom and top electrode layers of the capacitive sensors. Moreover, the dielectric layer was screen-patterned using a metal screen mask. RRAMs consisting of layers of patterned PI, Au/Al, titanium dioxide (TiO₂), Al/Au, and the final SU8 encapsulation layer were fabricated on a SiO₂ wafer. A thermal release tape was used to detach PRAMs from the SiO₂ wafer and transfer them onto a PDMS substrate. The sinter-free ink was used at the end to pattern the interconnects by means of screen printing. A similar fabrication method was used to fabricate QLEDs. The fabrication process consisted of patterning islands of PI, Ag/Au, zinc oxide (ZnO) nanoparticles, and red-emitting cadmium selenide/cadmium zinc sulphide (CdSe/CdZnS) quantum dots on a SiO₂ wafer. The islands were transferred onto a PDMS substrate by using a thermal release tape. This step was followed by thermal evaporation of a hole-transport layer (HTL), a hole-injection layer (HIL), an Ag layer as the anode, and a final SiO₂ encapsulation layer. The interconnects were then screen-printed. Figure 7B shows the system receiving four types of patterned stimuli (“S”, “N”, “U”, and “K”) through a 5 × 5 capacitive pressure sensor array. The output of the capacitive sensor array with and without 25% stretching is depicted in Figure 7C. The outputs of 25 capacitive sensors were converted to voltage signals and applied to the four RRAM arrays, shaping an artificial neural network (ANN). Two different neural networks were employed and trained for un-stretched and 25%-stretched RRAM synapses to classify the input patterns. The output results were then applied to the 5 × 5 QLED arrays to visualise the “S”, “N”, “U”, and “K” patterns formed by the QLED [Figure 7D]. The bioinspired SSNS can be used in applications such as smart skin prosthetic systems.

Figure 7. (A) Photograph of a stretchable sensory-neuromorphic system; (B) schematic illustration and image of SSNS and its overall operation; (C) image patterns (“S”, “N”, “U”, and “K”) provided by the pristine (top) and 25%-stretched (bottom) mechanoreceptors; and (D) photographs of epidermal photonic actuator array visualising the trained/inferred patterns of “S”, “N”, “U”, and “K”. This figure is quoted with permission from Kim et al.^[120]. QLED: Quantum dot light-emitting diode.

Pulse and respiration are two vital signals that provide an accurate assessment of individual health status^[140]. For effective detection of these subtle signals through the skin, soft and flexible wearable devices are required that could mimic the electrical properties and softness of human skin.

Wang et al. developed a flexible and wearable 3 × 3 pulse monitoring pressure sensor array that can acquire 3D pulse signals at three pulse positions called Chi, Cun, and Guan on the radial artery [Figure 8A]^[121]. These positions are important in palpation-based pulse diagnosis in traditional Chinese medical science (TCMS). The array was composed of resistive pressure sensors that converted the pulse signals of the artery to resistance changes, and the results were fed to a data acquisition system. A pre-processing circuit converted the resistance signals to voltage signals and amplified them; the amplified signals were subjected to noise filtering to remove high-frequency noise. Then, ADC conversion was applied to the multichannel pulse signals to construct a 3D map based on a polynomial fitting equation. A schematic illustration of this pulse sensing platform is presented in Figure 8A. The IDEs of each pressure sensor were screen-printed on a PET substrate by using a conductive Ag paste. A PDMS mould was prepared using a 3D-printed mould to define the sensing layers of the sensors above the IDEs. The PDMS mould and PET layer were bonded, and an ion gel layer was coated as the sensing layer through the PDMS mould above the electrodes. The device was then encapsulated with an ecoflex layer. The pulse signals (i.e., pulse strength versus time) recorded using nine pressure sensors are depicted in Figure 8B. The distance between two valleys (i.e., shown in yellow) defines the response time for one cardiac cycle. The three peaks in each cycle present the early systolic peak (P₁), inflexion point (P₂), and dicrotic peak (P₃). The similar peaks and valleys of all the waveforms indicate good synchronisation in the multichannel acquisition system. This pulse monitoring device can modernise Chinese medical science or be used in intelligent healthcare applications.

Figure 8. (A) Schematic illustration of the pulse-sensing platform comprising sensor arrays, signal processing, signal display, and 3D signal fitting; and (B) corresponding pulse signals (Cun, Guan, and Chi) detected by the nine sensor arrays^[121].

A triboelectric all-textile sensor array (TATSA) developed by Fan et al. can be stitched to clothes as a wristband, fingerstall, sock, and chest strap^[122]. TATSA was used to monitor arterial pulse waves and respiratory signals from various parts of the body. The physiological signals by TATSA recorded were amplified, low-pass-filtered, passed through an ADC converter, and transmitted wirelessly through Bluetooth to a mobile application for health status analysis [Figure 9A]. TATSA was knitted with conductive and commercial nylon yarns to prepare a full cardigan stitch [Figure 9B]. To prepare the conductive yarn, stainless-steel wires were twisted with several commercial terylene yarns. When an external force was applied and the two yarns came into contact with each other, equivalent charges with opposite polarities were generated on the surface of the two yarns because of triboelectrification. Once the two yarns separated, positive charges were induced in the inner stainless steel because of the electrostatic induction effect. The process of charge transfer during this contact–separation movement is presented in Figure 9C. The pulse waves recorded from the TATSA positioned at various points on the body of a 25-year-old woman are shown in Figure 9D. Three distinguishable peaks (P₁ to P₃) are visible on the pulse waveforms recorded at the neck, wrist, and fingertip positions, but the ankle waveform does not exhibit the P₂ peak. Figure 9E shows the positioning of TATSA on the chest for monitoring respiration. The detected waveform is the superimposition of the respiratory and heartbeat waveforms. A 0.8 Hz low-pass filter and a 0.8-20 Hz band-pass filter were used to separate the signals, as shown in Figure 9E. The sensor array could be used for long-term and non-invasive assessments of cardiovascular disease and sleep apnoea syndrome.

Figure 9. (A) Two TATSAs integrated into a shirt for real-time monitoring of pulse and respiratory signals; (B) schematic illustration of the combination of TATSA and clothes; (C) TATSA with the front, right, and top sides of the knit loops, and schematic illustrations of charge transfer through a contact unit during the pressure application and release steps; (D) photographs of TATSAs stitched into a wristband, fingerstall, sock, and chest strap, and the related measured waveforms; and (E) TATSA placed on the chest for measuring respiration signals, and the detected voltage-time and its decomposition into the heartbeat and respiratory waveforms^[122]. TATSA: Triboelectric all-textile sensor array.

Soft tongue stimulation electrode and tasting arrays

A tongue is a soft, flexible, and sensitive muscular organ with taste buds that include thousands of mechanical and taste receptors, as well as ion channels^[141]. The development of flexible electrode arrays for electrical stimulation of the tongue has facilitated investigation into a brain-tongue-machine interface from the perspective of the development of clinical neuromodulation and tissue-activation techniques. The five tastes experienced by mammals and humans are sweet, bitter, sour, salty, and umami. The recognition of sweet, bitter, and umami tastes is generally attributed to taste receptor cells (TRCs), whereas the recognition of sour and salty tastes is attributed to specific membrane channels on TRCs. This indicates that the sensing intensities of all five tastes rely on the distribution of tastebuds rather than the taste map shown in Figure 10A. In addition to chemically induced tastes, electrical stimulation can induce taste sensations.

Figure 10. (A) Schematics of electrical and chemical activation of some senses on the tongue (left) to determine whether the taste was related to the distribution of taste bud zones (middle), and the FTEAS device developed to electrically stimulate the tongue and detect activated cell-firing-related signals from each zone of the tongue simultaneously; (B) exploded-view schematic of FTEAS, and schematic of FTEAS showing FTEA, PLVAPS, and the data-acquisition module; (C) schematic illustration of the rat in vivo experiments, spike signals recorded from eight zones of the rat’s tongue, and zone-distribution radar map of the spike counts induced by chemical and electrical stimulations. This figure is quoted with permission from Huang et al.^[123]. FTEAS: Flexible tongue electrode array system; PLVAPS: portable low-voltage adjustable power supply.

Huang et al. developed a flexible tongue electrode array system (FTEAS) to electrically stimulate different regions of the tongue and collect multisite signals from the tongue [Figure 10A]^[123]. Figure 10B shows the layers of the FTEAS and a portable low-voltage adjustable power supply (PLVAPS)-stimulating module that was fabricated on a custom-designed PCB. The data acquisition module was responsible for collecting electrical signals from the FTEAS. The main power supply of the system was a 5 V DC battery. To fabricate the FTEAS, first, the Au/Ni/Cu electrodes and wiring were patterned onto a thin PI substrate. Then, another PI film with small holes above the Au/Ni/Cu electrodes was laminated to encapsulate the device. The results of simultaneous electrical stimulation of eight zones of a rat tongue are shown in Figure 10C. The recorded spike signal counts in each zone were summarised as the radar map. All spikes were normalised to the maximum counts recorded in Zone 4, which was induced by the 1 V, 10 Hz square wave stimulation signal. A similar number of spikes were observed for each chemical taste stimulation in all zones. The sinusoidal wave-induced spikes showed divergence in terms of spike counts in Zones 0 and 1. By contrast, symmetric spikes were obtained by the square-wave stimulations, similar to the case of chemical taste stimulations. Non-invasive electrotherapy for neurological diseases through the tongue and a tongue-machine interface is a potential application of FTEAS.

The tactile receptors (i.e., low-threshold mechanoreceptors) on the surface of the tongue are responsive to stimuli evoked by physical contact between the tip of the tongue, food, tooth, and lip^[142]. A few research studies have focused on oral somatosensation, which has not been understood very well thus far. For instance, an array of flexible electrode arrays has been developed to stimulate the dorsal surface of the tongue^[143]. This array can be used to evoke lost senses, such as vision, audition, proprioception, and the vestibular senses.

Inspired by the astringency perception mechanism of the human tongue upon licking a liquid [Figure 11A], Yeom et al. developed a soft hydrogel-based artificial tongue that mimics the thin salivary layer of the tongue^[124]. Unlike the previously developed artificial tongues, which require large amounts of liquid for operation, trace amounts of analytes are sufficient for detection with this device. This artificial tongue consists of a 3 × 3 sensor array with a saliva-like hydrogel acting as the sensing layer. The hydrogel consists of mucin as a secreted protein, lithium chloride (LiCl) as an electrolyte, and polyacrylamide (PAAm) as the 3D porous polymer network to facilitate facile electrolyte flow, as illustrated in Figure 11B. Additionally, a tannic acid (TA) solution was used to prepare the hydrogel layer because it could induce the formation of hydrophobic aggregates and enhance ionic conductivity. Direct UV polymerisation of the hydrogel on a poly(ethylene naphthalate) (PEN) substrate with patterned Au IDEs was realised using 3-(trimethoxysilyl)propyl methacrylate (TMSPMA) as the chemical anchoring agent. When 0.1 and 1 wt.% TA solutions were applied at five different spots on the array, the sensor measured the corresponding current changes to generate taste maps [Figure 11C]. The largest current changes were observed at the spots where 1 wt.% solution was applied, whereas the spots where 0.1 wt.% solution was applied exhibited smaller current changes. This artificial tongue could be used as a platform for future applications involving portable taste-monitoring devices and humanoid robots.

Figure 11. (A) Schematic illustration of wipe-and-detection of the human tongue; (B) an image of the artificial tongue and schematic illustration of the astringency-sensing principle of the artificial tongue; and (C) application of five drops of 1 wt.% TA to the artificial tongue array (left panel) and two drops of 0.1 and 1 wt.% TA to the array (right panel), along with the corresponding taste mapping data^[124]. TA: Tannic acid.

Soft odour identification sensor arrays

The human olfactory system is responsible for our sense of smell. The nose houses various receptor families that detect volatile chemical stimuli which dissolve in the mucus lining^[144]. The action potentials generated in the receptors travel along their axons, which terminate in the olfactory bulb. The olfactory information then travels to the olfactory cortex for processing, as shown in Figure 12A^[145]. The electric nose (e-nose) that could mimic this operation could include an array of chemical sensors, electronic circuitry (e.g., sampling, filtering, and signal conditioning), and data analysis software [Figure 12A]^[145].

Figure 12. (A) Schematic illustration comparing the mammalian olfactory system with an electronic nose (e-nose)^[145]; (B) Schematic illustration of wearable e-nose with the ZigBee module and a photo of the fabricated e-nose sensor array; and (C) sensor array exposed to 200 ppm of volatiles, namely ammonia, acetic acid, acetone, and ethanol. This figure is quoted with permission from Lorwongtragool et al.^[125].

Lorwongtragool et al. developed a wearable e-nose by integrating a 4 × 2 chemical sensor array with a Zigbee wireless communication system [Figure 12B]^[125]. The sensor array comprised eight fully inkjet-printed gas sensors on a PEN film with Ag ink IDEs. The resistance of each sensor was determined by connecting it to a voltage divider. The output voltages of eight sensors were then applied to an eight-channel analogue multiplexer connected to a USB data acquisition (DAQ) device. To fabricate the sensing layers of sensors 1-4, layers of a multi-walled CNT (MWCNT) solution were inkjet-printed on electrodes, followed by the printing of a polymer layer composed of one of the following polymers: polyvinyl chloride (PVC), poly(styrene-co-maleic anhydride), cumene terminated poly(styrene-co-maleic anhydride) (Cumene-PSMA), polysulphone (PSE), and polyvinylpyrrolidone (PVP). For sensors 5-8, each polymer was blended with the MWCNT solution and printed on electrodes. The sensor array was exposed to four individual volatile organic compounds (VOCs) present in the volatiles released from the axillary skin. The sensor response percentages for the four VOCs are shown in Figure 12C. The e-nose was used to monitor the body odour of a participant before exercise, after exercise, and after relaxing for a long period post-exercise. The data were collected for 5 min in each step, and the entire data-collection process was repeated thrice. Principal component analysis (PCA) based on unsupervised learning was used to observe the odour progression in each minute and classify the odour of the armpit when performing different activities. This wearable e-nose could be used to detect unusual odours corresponding to various health statuses.

Zheng et al. developed a skin odour detection sensor array (i.e., e-nose system) consisting of 3 × 2 sensors based on CNTs and polymer suspension solutions as the sensing materials^[126]. This flexible sensor array data was read using a voltage divider circuit and interfaced with an MCU for data acquisition. A Bluetooth chip and a button cell were used for wireless transmission and as the power supply, respectively [Figure 13A]. To prepare the sensing material solutions, one of the three polymers [hydroxypropyl methylcellulose (HPMC), poly(methyl vinyl ether-co-maleic anhydride) (PMVEMA), and polyvinyl pyrrolidone (PVP)] was mixed with carboxylated or hydroxylated CNTs by ball milling. Ag IDEs were screen-printed on a PET substrate. Then, 5 μL of the sensing solution was drop-casted on the respective electrodes and dried. To avoid direct contact with sweat, a hydrophobic and breathable PVDF membrane was laminated onto the sensor array. A schematic of the wearable e-nose system integrated into an armband is shown in Figure 13B. The sensor was placed under the armpit, while the related circuitry was placed on the bicep. The responses of each sensor to 5-25 ppm of hexanoic acid, dodecane, and decanal (i.e., elements with an odour similar to sweat) were measured to calculate their sensitivities [Figure 13C]. The e-nose system was tested on eight participants while they walked for 63 min on a treadmill. The experiment was repeated on four consecutive days. The responses of the sensors were then used for feature extraction, and a K-means clustering algorithm was applied to identify human odour. This e-nose system that can detect human skin odour can be used for individual identification or to determine individual physiological status.

Figure 13. (A) Schematic illustration of the working principle of the wearable e-nose system; (B) photograph of a wearable e-nose on the armpit of a subject, with an integrated chemical sensor array and wireless circuit; and (C) responses of the sensor array to 5-25 ppm of hexanoic acid, dodecane, and decanal. This figure is quoted with permission from Zheng et al.^[126]. MCU: Micro-controller unit; PVDF: poly(vinylidene fluoride); VOCs: volatile organic compounds.

Soft sound sensor arrays

Sound energy is conveyed by the pinna and auditory canal to the eardrum^[146,147]. Vibrations of the eardrum are amplified and transmitted by the middle ear bones, namely, the malleus, incus, and stapes, to the inner ear fluid. The cochlea is a fluid-filled duct in the inner ear that is responsible for converting sound vibrations into electrical signals with the help of specialised sensory cells. These electrical signals are then transmitted to the brain through the auditory nerve for processing. The structure of the human ear is shown in Figure 14A^[146].

Figure 14. (A) Structure of the human ear^[146]; (B) Nine different modules of the FPC and an optical image of the PTAS; (C) piezoelectric voltage distribution of the array on the 3D ear, and 3D sound diagrams of the cases in which the sound source was placed above pixel 5 (top) and pixel 8 (bottom). This figure is quoted with permission from Yang et al.^[127]. PTAS: P(VDF-TrFE) acoustic sensor; PVDF-TrFE: poly(vinylidene fluoride-trifluoroethylene).

Yang et al. developed a 3 × 3 array of self-powered acoustic sensors based on a piezoelectric poly(vinylidene fluoride-trifluoroethylene) [P(VDF-TrFE)] thin film (PTF) [i.e., P(VDF-TrFE) acoustic sensor (PTAS) array] that was integrated with a 3D-printed bionic ear^[127]. The PTF layer converted the mechanical vibration of sound waves to electric signals, similar to the eardrum and cochlea. Data-acquisition cards and circuits were developed to collect and transmit electrical signals to the processing computer, similar to the nerve cells. To fabricate the PTAS array, P(VDF-TrFE) powder was dissolved in methyl ethyl ketone, and the solution was stirred magnetically. Thereafter, the solution was coated on an FPCB and placed in a vacuum chamber to form a PTF film, which was then annealed. The annealed PTF film was then polarised in situ. Afterwards, a Cu foil layer was pasted on top of the film. Finally, the resulting flexible circuit was attached to a 3D-printed ear, as shown in Figure 14B. The 3D-printed ear was used to map sound waves from two different sound sources at two different locations [Figure 14C]. The piezoelectric voltage of each pixel was then collected, and a corresponding colour-scaled 3D diagram was simulated based on the acquired piezoelectric signals. When the sound source was directly opposite the 3D-printed ear, pixel 5 exhibited the highest piezoelectric response of 6.232 mV because it was the nearest to the sound source. When the sound source was located diagonally opposite to the 3D-printed ear, pixel 8 exhibited the highest piezoelectric voltage of 5.612 mV because it was the nearest to the sound source. This PTAS array can be used for sound identification and localisation in bionic applications.